TSMC (TSM) Expansion Speed Is Top AI Bubble Indicator, Atreides' Baker Says, Touting Amazon (AMZN) Trainium

Atreides Management CIO Gavin Baker said on May 27, 2026, that Taiwan Semiconductor Manufacturing Co.'s (TSMC) capacity decisions are the single most important metric for judging whether an AI bubble will form. Speaking at the Sohn Conference, Baker argued TSMC acts as a gatekeeper, balancing expansion to maintain its technology lead while preventing chip oversupply that could burst a valuation bubble. If TSMC fully met Nvidia's (NVDA) demands, he estimated Nvidia could sell $2 trillion to $3 trillion in GPUs in 2026-2027, risking a glut. Baker also flagged Amazon's (AMZN) Trainium chip as the most underrated AI play, predicting its Trainium 3 launch in the second half of 2026 will challenge Nvidia's dominance. He highlighted Anthropic's surging revenue, adding $11 billion in monthly annualized recurring revenue. Memory prices have risen 60-70% this year, but Baker cautioned that an algorithmic breakthrough reducing the need for massive compute would be the real risk to AI infrastructure bets.